$30 a Month for AI: An Indie Hacker’s Receipt

2026-10-26 Spend less EN

01 / Your $30 AI envelope, before the first request

A solo project does not need an unlimited AI subscription by default. It needs a spending ceiling and a clear answer to one question: what work deserves a paid model call? Start with a $30 monthly envelope for coding assistance, product experiments, and a small amount of production traffic. This is a planning budget, not a promise that every application can run for $30.

CheapRelay fits this approach because it charges per real token, including cache reads and writes, with no monthly fee or subscription. One key can serve development tools and application requests. That removes a fixed subscription line, but it does not remove the need to measure usage.

02 / The receipt: $18 for planned work, $12 for uncertainty

Here is a sample monthly allocation. The token prices below are fictional planning assumptions, not CheapRelay price quotes or rates for any named model. Replace them with current prices before funding your account.

■ Routine work: 6 million input tokens at an assumed $1 per million, plus 2 million output tokens at $3 per million: $12.
■ Difficult work: 750,000 input tokens at an assumed $4 per million, plus 250,000 output tokens at $12 per million: $6.
■ Cache reads and writes: $4 spending allowance, calculated separately using the selected model’s actual cache rates.
■ Retries, failed experiments, and repeated tool loops: $5 spending allowance.
■ Unallocated reserve: $3.
■ Total monthly envelope: $30.

The $12 routine line is straightforward: $6 of input plus $6 of output. The $6 difficult-work line is $3 of input plus $3 of output. The remaining $12 is headroom, not an estimate of tokens consumed. If you do not spend it, leave it unspent.

03 / Translate the $12 routine line into actual jobs

Suppose one routine job uses 3,000 input tokens and produces 1,000 output tokens. Under the fictional routine rates above, that job costs $0.006: $0.003 for input and $0.003 for output. Two thousand such jobs consume the entire $12 allocation.

That is roughly 67 jobs a day over 30 days. A job might be a short classification, a draft response, or a focused code question. It is not necessarily one user action: an agent can make several model calls while completing a single task. Count requests and tokens, not just clicks.

Long conversations change the arithmetic. Repeatedly sending a large repository, chat history, or document bundle can exhaust the input allocation before you reach the planned job count. Measure the complete request, including instructions and tool results.

04 / Give expensive reasoning a $6 gate

Use the lower-cost route first for bounded tasks. Escalate when a response fails a test, misses a required field, or cannot resolve a specific ambiguity. Do not send every request through a more expensive route simply because it is available.

For example, compare DeepSeek V4 Flash and Claude Sonnet 5 using current prices and your own test cases before choosing their roles. The fictional rates in this receipt are not assigned to either model. A sensible choice depends on both the billed cost and whether the result passes your checks.

05 / Stop small mistakes before they become a $30 afternoon

■ Record input, output, cache usage, and billed cost for each request.
■ Review spending at $10 and $20; pause nonessential work at $27.
■ Set application-side request limits and maximum output lengths.
■ Allow one automatic retry, then surface the failure instead of looping.
■ Use a separate project key where supported, and never expose credentials in browser code.

Check whether your provider offers enforceable spending limits. Application alerts alone are not a billing cap. For OpenAI-compatible clients, CheapRelay uses https://api.cheapapirelay.com/v1; existing SDK integrations can use that base URL.

06 / Buy a measured month, not a bigger commitment

Sign up for CheapRelay and begin with a small, measured workload. After seven days, compare actual spending with this receipt and revise the allocations. Keep the $30 ceiling until real usage justifies raising it. The goal is not to spend the entire budget; it is to ship useful work without collecting another recurring bill.

More reading

Date	Aisle	Written in	Title
2026-10-26	Spend less	EN	$3.45 for 95 accepted answers beats cheap tokens A worked receipt shows why token prices can mislead: count retries, rejected outputs and cache charges before ranking models.
2026-10-26	Spend less	EN	The $0.004 AI Task With a $0.0566 Agent Receipt A hypothetical agent receipt shows how context, retries, evaluation and logging turn a $0.004 task into $0.0566 per accepted result.
2026-10-26	Spend less	EN	$2.53 for 10,000 JSON Jobs: Read the Receipt A worked JSON-output budget: token costs, schema overhead, retries, and the checks that keep cheap API calls from becoming expensive workflows.
2026-10-26	Spend less	EN	$0.012 Either Way: Streaming Changes the Habit A token-by-token receipt shows why streaming has no automatic discount, but stopping early and avoiding retries can change your API bill.
2026-10-26	Spend less	EN	$1.71 to $1.02: A Long-Chat Summarisation Receipt A worked token receipt shows when summarising chat history saves money, including the summary fee, break-even point and caching caveats.
2026-10-26	Spend less	EN	¥0.32 for a 100,000-token translation batch A worked translation receipt shows how to rank model quotes by yuan spent, then adjust for retries, cache charges and human review.